k8s.go

/*
Copyright 2023 The Radius Authors.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

	http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
package validation

import (
	"context"
	"fmt"
	"io"
	"log"
	"os"
	"testing"
	"time"

	kuberneteskeys "github.com/radius-project/radius/pkg/kubernetes"
	"github.com/stretchr/testify/assert"
	"github.com/stretchr/testify/require"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/api/meta"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
	"k8s.io/apimachinery/pkg/labels"
	"k8s.io/apimachinery/pkg/runtime/schema"
	watchk8s "k8s.io/apimachinery/pkg/watch"
	"k8s.io/client-go/discovery/cached/memory"
	"k8s.io/client-go/dynamic"
	"k8s.io/client-go/kubernetes"
	v1 "k8s.io/client-go/kubernetes/typed/core/v1"
	"k8s.io/client-go/restmapper"
)

const (
	IntervalForDeploymentCreation = 3 * time.Second
	IntervalForPodShutdown        = 3 * time.Second
	IntervalForResourceCreation   = 3 * time.Second

	// We want to make sure to produce some output any time we're in a watch
	// otherwise it's hard to know if it got stuck.
	IntervalForWatchHeartbeat = 10 * time.Second
)

type K8sObjectSet struct {
	Namespaces map[string][]K8sObject
}

type K8sObject struct {
	GroupVersionResource schema.GroupVersionResource
	Labels               map[string]string
	Kind                 string
	SkipLabelValidation  bool
	ResourceName         string
}

// NewK8sPodForResource creates a new K8sObject with Kind set to "Pod" and the selector labels for the pod
// set to the given application and name.
func NewK8sPodForResource(application string, name string) K8sObject {
	return K8sObject{
		// NOTE: we use the selector labels here because the selector labels are intended
		// to be determininistic. We might add things to the descriptive labels that are NON deterministic.
		GroupVersionResource: schema.GroupVersionResource{
			Group:    "",
			Version:  "v1",
			Resource: "pods",
		},
		Kind:   "Pod",
		Labels: kuberneteskeys.MakeSelectorLabels(application, name),
	}
}

// ValidateLabels creates a copy of the K8sObject and sets the SkipLabelValidation field based on the validate parameter.
func (k K8sObject) ValidateLabels(validate bool) K8sObject {
	copy := k
	copy.SkipLabelValidation = !validate
	return copy
}

// NewK8sHTTPProxyForResource creates a K8sObject for a HttpProxy with the Labels set to the application and name provided.
func NewK8sHTTPProxyForResource(application string, name string) K8sObject {
	return K8sObject{
		GroupVersionResource: schema.GroupVersionResource{
			Group:    "projectcontour.io",
			Version:  "v1",
			Resource: "httpproxies",
		},
		Kind:   "HTTPProxy",
		Labels: kuberneteskeys.MakeSelectorLabels(application, name),
	}
}

// NewK8sServiceForResource creates a new K8sObject for a service with the Labels set to the application and name.
func NewK8sServiceForResource(application string, name string) K8sObject {
	return K8sObject{
		GroupVersionResource: schema.GroupVersionResource{
			Group:    "",
			Version:  "v1",
			Resource: "services",
		},
		Kind:   "Service",
		Labels: kuberneteskeys.MakeSelectorLabels(application, name),
	}
}

// NewK8sSecretForResource creates a K8sObject for a secret with the Labels set to the application and name.
func NewK8sSecretForResource(application string, name string) K8sObject {
	return K8sObject{
		GroupVersionResource: schema.GroupVersionResource{
			Group:    "",
			Version:  "v1",
			Resource: "secrets",
		},
		Kind:   "Secret",
		Labels: kuberneteskeys.MakeSelectorLabels(application, name),
	}
}

// NewK8sSecretForResourceWithResourceName creates a K8sObject for a secret with the Labels set to the application and name.
func NewK8sSecretForResourceWithResourceName(resourceName string) K8sObject {
	return K8sObject{
		GroupVersionResource: schema.GroupVersionResource{
			Group:    "",
			Version:  "v1",
			Resource: "secrets",
		},
		Kind:         "Secret",
		ResourceName: resourceName,
	}
}

// SaveContainerLogs watches for all pods in the given namespace and saves their container logs to disk.
func SaveContainerLogs(ctx context.Context, k8s *kubernetes.Clientset, namespace string, logPrefix string) (watchk8s.Interface, error) {
	return watchForPods(ctx, k8s, namespace, logPrefix, "")
}

func watchForPods(ctx context.Context, k8s *kubernetes.Clientset, namespace string, logPrefix string, labelSelector string) (watchk8s.Interface, error) {
	if err := os.MkdirAll(logPrefix, os.ModePerm); err != nil {
		log.Printf("Failed to create output log directory '%s' Error was: '%q'. Container logs will be discarded", logPrefix, err)
		return nil, nil
	}

	podClient := k8s.CoreV1().Pods(namespace)

	// Filter only Radius Applications for a pod
	podList, err := podClient.Watch(ctx, metav1.ListOptions{
		Watch:         true,
		LabelSelector: labelSelector,
	})
	if err != nil {
		return nil, err
	}

	go func() {
		pods := map[string]bool{}
		for event := range podList.ResultChan() {
			pod, ok := event.Object.(*corev1.Pod)
			if !ok {
				_, ok := event.Object.(*metav1.Status)
				if ok {
					// Ignore statuses, these might be the result of a connection dropping or the watch being cancelled.
					continue
				}

				log.Printf("Could not convert object to pod or status, was %+v.", event.Object)
				continue
			}

			// Skip streaming log when Pod is in pending state.
			if pod.Status.Phase == corev1.PodPending {
				continue
			}

			log.Printf("Start streaming Kubernetes logs - Pod %s is in state: %s", pod.Name, pod.Status.Phase)

			// Only start one log capture per pod
			_, ok = pods[pod.Name]
			if ok {
				continue
			}
			pods[pod.Name] = true

			for _, container := range pod.Spec.Containers {
				go streamLogFile(ctx, podClient, *pod, container, logPrefix)
			}
		}
	}()

	return podList, nil
}

// See https://github.com/dapr/dapr/blob/22bb68bc89a86fc64c2c27dfd219ba68a38fb2ad/tests/platforms/kubernetes/appmanager.go#L706 for reference.
func streamLogFile(ctx context.Context, podClient v1.PodInterface, pod corev1.Pod, container corev1.Container, logPrefix string) {
	filename := fmt.Sprintf("%s/%s.%s.log", logPrefix, pod.Name, container.Name)
	log.Printf("Streaming Kubernetes logs to %s", filename)
	req := podClient.GetLogs(pod.Name, &corev1.PodLogOptions{
		Container: container.Name,
		Follow:    true,
	})
	stream, err := req.Stream(ctx)
	if err != nil && err == ctx.Err() {
		return
	} else if err != nil {
		log.Printf("Error reading log stream for %s. Error was %+v", filename, err)
		return
	}
	defer stream.Close()

	fh, err := os.Create(filename)
	if err != nil {
		log.Printf("Error creating %s. Error was %s", filename, err)
		return
	}
	defer fh.Close()

	buf := make([]byte, 2000)

	for {
		numBytes, err := stream.Read(buf)

		if err == io.EOF {
			break
		}

		if err != nil && err == ctx.Err() {
			return
		} else if err != nil {
			log.Printf("Error reading log stream for %s. Error was %+v", filename, err)
			return
		}

		if numBytes == 0 {
			continue
		}

		_, err = fh.Write(buf[:numBytes])
		if err != nil {
			log.Printf("Error writing to %s. Error was %s", filename, err)
			return
		}
	}

	log.Printf("Saved container logs to %s", filename)
}

// ValidateObjectsRunning checks if the expected Kubernetes objects are running in the given namespace.
func ValidateObjectsRunning(ctx context.Context, t *testing.T, k8s *kubernetes.Clientset, dynamic dynamic.Interface, expected K8sObjectSet) {
	restMapper := restmapper.NewDeferredDiscoveryRESTMapper(memory.NewMemCacheClient(k8s.DiscoveryClient))
	for namespace, expectedObjects := range expected.Namespaces {
		log.Printf("validating objects in namespace %v", namespace)
		namespaceTypes := map[schema.GroupVersionResource][]K8sObject{}
		for _, obj := range expectedObjects {
			_, ok := namespaceTypes[obj.GroupVersionResource]
			if ok {
				namespaceTypes[obj.GroupVersionResource] = append(namespaceTypes[obj.GroupVersionResource], obj)
			} else {
				namespaceTypes[obj.GroupVersionResource] = []K8sObject{obj}
			}
		}
		for {
			validated := true
			select {
			case <-time.After(IntervalForResourceCreation):
				for resourceGVR, expectedInNamespace := range namespaceTypes {
					r, err := restMapper.KindFor(resourceGVR)
					assert.NoErrorf(t, err, "failed to get kind for %s", resourceGVR)

					mapping, err := restMapper.RESTMapping(r.GroupKind(), r.Version)
					assert.NoErrorf(t, err, "failed to get rest mapping for %s", r.GroupKind())

					var deployedResources *unstructured.UnstructuredList
					if mapping.Scope == meta.RESTScopeNamespace {
						deployedResources, err = dynamic.Resource(mapping.Resource).Namespace(namespace).List(ctx, metav1.ListOptions{})
					} else {
						deployedResources, err = dynamic.Resource(mapping.Resource).List(ctx, metav1.ListOptions{})
					}
					assert.NoErrorf(t, err, "could not list deployed resources of type %s in namespace %s", resourceGVR.GroupResource(), namespace)
					t.Logf("Listed %d deployed resources of type %s in namespace %s", len(deployedResources.Items), resourceGVR.GroupResource(), namespace)
					validated = validated && matchesActualLabels(t, expectedInNamespace, deployedResources.Items)

				}
			case <-ctx.Done():
				assert.Fail(t, "timed out after waiting for services to be created")
				return
			}

			if validated {
				break
			}
		}

		// All of the resources have been created but we want to check conditions as well
		for resourceGVR, expectedInNamespace := range namespaceTypes {
			if resourceGVR.Resource != "pods" {
				continue
			}

			for _, selector := range expectedInNamespace {
				t.Logf("Checking pods in %s with %s", namespace, selector.Labels)
				actualPods, err := k8s.CoreV1().Pods(namespace).List(ctx, metav1.ListOptions{
					LabelSelector: labels.SelectorFromSet(selector.Labels).String(),
				})
				assert.NoErrorf(t, err, "failed to list pods in namespace %s", namespace)

				for _, pod := range actualPods.Items {
					t.Logf("Checking pod: %s:%s", pod.Namespace, pod.Name)
					monitor := PodMonitor{K8s: k8s, Pod: pod}
					monitor.ValidateRunning(ctx, t)
					t.Logf("Pod is ready %s:%s", pod.Namespace, pod.Name)
				}
			}
		}
	}
}

// ValidateNoPodsInApplication checks if there are any pods in the given namespace for the given application and waits for
// them to be deleted if found.
func ValidateNoPodsInApplication(ctx context.Context, t *testing.T, k8s *kubernetes.Clientset, namespace string, application string) {
	labelset := kuberneteskeys.MakeSelectorLabels(application, "")

	actualPods, err := k8s.CoreV1().Pods(namespace).List(context.Background(), metav1.ListOptions{
		LabelSelector: labels.SelectorFromSet(labelset).String(),
	})
	assert.NoErrorf(t, err, "failed to list pods in namespace %s", namespace)

	logPods(t, actualPods.Items)

	// There's an inherent race condition in verifying that the Pods are gone. We're at the
	// mercy of the Kubernetes event loop. We'll wait for pods to disappear if we find them.
	if len(actualPods.Items) == 0 {
		return
	}
	for {
		select {

		case <-ctx.Done():
			assert.Fail(t, "timed out waiting for pods to be deleted")
			return

		case <-time.After(IntervalForPodShutdown):
			t.Logf("at %s waiting for pods in namespace %s for application %s to shut down.. ", time.Now().Format("2006-01-02 15:04:05"), namespace, application)

			actualPods, err := listPodsWithRetries(t, k8s, labelset, namespace, application)
			assert.NoError(t, err)

			logPods(t, actualPods.Items)
			if len(actualPods.Items) == 0 {
				t.Logf("success! no pods found in namespace %s for application %s", namespace, application)
				return
			}
		}
	}
}

func listPodsWithRetries(t *testing.T, k8s *kubernetes.Clientset, labelset map[string]string, namespace, application string) (*corev1.PodList, error) {
	// Need to retry because of AKS error: https://github.com/radius-project/radius/issues/2484
	retries := 3
	for i := 1; i <= retries; i++ {
		actualPods, err := k8s.CoreV1().Pods(namespace).List(context.Background(), metav1.ListOptions{
			LabelSelector: labels.SelectorFromSet(labelset).String(),
		})
		if err == nil {
			return actualPods, nil
		}

		t.Logf("failed to list pods in namespace %s for application %s. retrying in %fs... (%d/%d)", namespace, application, IntervalForPodShutdown.Seconds(), i, retries)
		time.Sleep(IntervalForPodShutdown)
	}

	return nil, fmt.Errorf("failed to list pods in namespace %s for application %s after %d retries", namespace, application, retries)
}

type PodMonitor struct {
	K8s *kubernetes.Clientset
	Pod corev1.Pod
}

// PodMonitor ValidateRunning watches a pod for its status to become running and checks its readiness, retrying a few times
// if the readiness check fails. If the pod enters a failing state, an error is returned.
func (pm PodMonitor) ValidateRunning(ctx context.Context, t *testing.T) {
	if pm.Pod.Status.Phase == corev1.PodRunning {
		if checkReadiness(&pm.Pod) {
			return
		}
	}

	t.Logf("watching pod %v for status.. current: %v", pm.Pod.Name, pm.Pod.Status)

	// If we're not in the running state, we need to wait a bit to see if things work out.
	watch, err := pm.K8s.CoreV1().Pods(pm.Pod.Namespace).Watch(ctx, metav1.SingleObject(pm.Pod.ObjectMeta))
	require.NoErrorf(t, err, "failed to watch pod: %v", pm.Pod.Name)
	defer watch.Stop()

	// Sometimes, the pods may take a little bit to become ready.
	// Therefore, if the readiness check fails, will retry a few times instead
	// of instantly failing
	const MaxRetryAttempts = 10
	attempt := 0
	for {
		select {
		case <-time.After(IntervalForWatchHeartbeat):
			t.Logf("watching pod %v for status.. current: %v", pm.Pod.Name, pm.Pod.Status)

		case event := <-watch.ResultChan():
			pod, ok := event.Object.(*corev1.Pod)
			if !ok {
				// Check the status if there is a failure.
				// Errors usually have a status as the object type
				if event.Type == watchk8s.Error {
					status, ok := event.Object.(*metav1.Status)
					if ok {
						if status.Reason == "Expired" {
							t.Logf("skipped pod watch expiration error: %s", status.Message)
							continue
						}
						require.Fail(t, fmt.Sprintf("pod watch error with status reason: %s, message: %s", status.Reason, status.Message))
					}
					require.Fail(t, fmt.Sprintf("object %T is not a status", event.Object))
				}
				require.Fail(t, fmt.Sprintf("object %T is not a pod, event type is %v", event.Object, event.Type))
			}

			if pod.Status.Phase == corev1.PodRunning {
				t.Logf("success! pod %v has status: %v", pod.Name, pod.Status)
				if checkReadiness(pod) {
					return
				}
				if attempt >= MaxRetryAttempts {
					assert.Failf(t, "pod failed readiness checks.", "Pod Name: %v", pod.Name)
				} else {
					t.Logf("Readiness check failed. Retrying - attempt %d", attempt)
					attempt++
				}
			} else if pod.Status.Phase == corev1.PodFailed {
				assert.Failf(t, "pod entered a failing state", "Pod Name: %v", pod.Name)
				return
			}

			t.Logf("watching pod %v for status.. current: %v", pod.Name, pod.Status)

		case <-ctx.Done():
			assert.Failf(t, "timed out after waiting for pod %v to enter running status", pm.Pod.Name)
			return
		}
	}
}

func checkReadiness(pod *corev1.Pod) bool {
	for _, condition := range pod.Status.Conditions {
		if condition.Type == corev1.ContainersReady &&
			condition.Status == corev1.ConditionTrue {
			// All okay
			return true
		}
	}
	return false
}

func logPods(t *testing.T, pods []corev1.Pod) {
	t.Log("Found the following pods:")
	if len(pods) == 0 {
		t.Logf("(none)")
	} else {
		for _, pod := range pods {
			t.Logf("namespace: %v name: %v labels: %v", pod.Namespace, pod.Name, pod.Labels)
		}
	}
}

func matchesActualLabels(t *testing.T, expectedResources []K8sObject, actualResources []unstructured.Unstructured) bool {
	remaining := []K8sObject{}

	for _, expectedResource := range expectedResources {
		if expectedResource.SkipLabelValidation && expectedResource.Kind != "Secret" {
			continue
		}
		resourceExists := false
		for idx, actualResource := range actualResources {
			if expectedResource.SkipLabelValidation {
				if actualResource.GetName() == expectedResource.ResourceName {
					resourceExists = true
					actualResources = append(actualResources[:idx], actualResources[idx+1:]...)
					break
				}
			} else {
				if labelsEqual(expectedResource.Labels, actualResource.GetLabels()) {
					resourceExists = true
					actualResources = append(actualResources[:idx], actualResources[idx+1:]...)
					break
				} else {
					t.Logf("Resource: %s Expected labels %v, got %v", actualResource.GetName(), expectedResource.Labels, actualResource.GetLabels())
				}
			}
		}
		if !resourceExists {
			remaining = append(remaining, expectedResource)
		}
	}

	for _, remainingResource := range remaining {
		log.Printf("Failed to validate resource of type %s with labels %s", remainingResource.GroupVersionResource.Resource, remainingResource.Labels)
	}
	return len(remaining) == 0
}

func labelsEqual(expectedLabels map[string]string, actualLabels map[string]string) bool {
	for key, value := range expectedLabels {

		if actualLabels[key] != value {
			return false
		}
	}
	return true
}

// NewDaprComponent creates a K8sObject for a Dapr component with the Labels set to the application and name.
func NewDaprComponent(application string, name string) K8sObject {
	return K8sObject{
		GroupVersionResource: schema.GroupVersionResource{
			Group:    "dapr.io",
			Version:  "v1alpha1",
			Resource: "components",
		},
		Kind:   "Component",
		Labels: kuberneteskeys.MakeSelectorLabels(application, name),
	}
}